Re: GPH: RE: Concurrency and Haskell
Murray, There are several parallel Haskell implementations: a survey of them has just appeared in Journal of Func. Prog Vols 4&5 (July & Sept 2002). Implementations are available for o Eden http://www.mathematik.uni-marburg.de/inf/eden/ o GpH http://www.cee.hw.ac.uk/~dsg/gph/ My group works on Glasgow parallel Haskell (GpH) which extends Haskell 98 with a parallel composition combinator. As Simon said the main implemntation, GUM, buys portability using the relatively slow PVM communications library. This doesn't matter so much on distributed memory machines, and we've recently achieved some quite respectable results SunServer shared-memory machines. Simon Marlowe developed an alternative SMP implementation of GpH a couple of years ago that may be more suitable for a Mosix platform, but I'm not sure of the status of that implementation now. Phil On Mon, 19 Aug 2002 18:04:44 +0100 Simon Peyton-Jones <simonpj@microsoft.com> wrote:
Hi Murray
I'm catching up with my email backlog, but I didn't see a reply to your message, so I thought I'd reply. The parallel-Haskell crew are at http://www.cee.hw.ac.uk/~dsg/gph/ There's a mailing list that I'm ccing.
GHC can be built to run on a shared-memory multiprocessor, but we have not built and tested that for some time.
Untrue - both Eden and GpH are currently used on SunServer SMPs.
GPH is aimed more at distributed-memory machines, and has quite a bit more code layered on top of GHC. Details on the GPH page above.
Let me know if I can help. It'd be great to get some more parallel-Haskell stuff going.
Simon
| A group of us at Brooklyn College (City University of New | York) are in the latter stages of setting up a Mosix cluster | that we would like to use for research on, among other | things, parallel execution of Haskell code and run-time | optimization of both automatic and programmed parallelization. | | I would greatly appreciate hearing from anyone who can | provide advice on | an appropriate bibliography on the language implementation and past | work in this area (other than the code in the Haskell compilers and | libraries, of course), approaches, ongoing work, etc. | | Also, I would be most interested in hearing from anyone | interested in joining our effort (no problems with remote | access to the cluster by way of SSH), or making arrangements | for cooperation in research in this area, although, of | course, access to our equipment will be subject to approval | from "higher authorities" (however, note that I really don't | expect difficulty here, since our administration has already | shown considerable interest in supporting the project). | | Replies to the list will be read, but those who wish may mail | me directly at mgross@dorsai.org. | | Thanks in advance for any replies. | | Murray Gross | Adj. Lecturer, Brooklyn College, | City University of New York | | | | _______________________________________________ | Haskell mailing list | Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell |
-------------------------------------------------- Phil Trinder Department of Computing and Electrical Engineering Heriot Watt University Riccarton Edinburgh, EH14 4AS E-mail: trinder@cee.hw.ac.uk Teleph: +44 (0)131 451 3435 Depart: +44 (0)131 451 3328 Fasmly: +44 (0)131 451 3327 Intrnt: http://www.cee.hw.ac.uk/~trinder
Speaking of GpH, I wonder how is GdH coming along? It seems that the installation instruction on http://www.cee.hw.ac.uk/~dsg/gdh/ is still incomplete... As far as I know, Mosix has nothing to do with PVM, so am I right to say that GpH still needs PVM even you use it on a Mosix cluster? Then what will Mosix provide when you can have a PVM cluster without Mosix running the same thing? Regards, .paul. On Tue, Aug 20, 2002 at 12:09:28PM +0100, Phil Trinder wrote:
Murray,
There are several parallel Haskell implementations: a survey of them has just appeared in Journal of Func. Prog Vols 4&5 (July & Sept 2002). Implementations are available for o Eden http://www.mathematik.uni-marburg.de/inf/eden/ o GpH http://www.cee.hw.ac.uk/~dsg/gph/
My group works on Glasgow parallel Haskell (GpH) which extends Haskell 98 with a parallel composition combinator. As Simon said the main implemntation, GUM, buys portability using the relatively slow PVM communications library. This doesn't matter so much on distributed memory machines, and we've recently achieved some quite respectable results SunServer shared-memory machines.
Simon Marlowe developed an alternative SMP implementation of GpH a couple of years ago that may be more suitable for a Mosix platform, but I'm not sure of the status of that implementation now.
Phil
On Mon, 19 Aug 2002 18:04:44 +0100 Simon Peyton-Jones <simonpj@microsoft.com> wrote:
Hi Murray
I'm catching up with my email backlog, but I didn't see a reply to your message, so I thought I'd reply. The parallel-Haskell crew are at http://www.cee.hw.ac.uk/~dsg/gph/ There's a mailing list that I'm ccing.
GHC can be built to run on a shared-memory multiprocessor, but we have not built and tested that for some time.
Untrue - both Eden and GpH are currently used on SunServer SMPs.
GPH is aimed more at distributed-memory machines, and has quite a bit more code layered on top of GHC. Details on the GPH page above.
Let me know if I can help. It'd be great to get some more parallel-Haskell stuff going.
Simon
| A group of us at Brooklyn College (City University of New | York) are in the latter stages of setting up a Mosix cluster | that we would like to use for research on, among other | things, parallel execution of Haskell code and run-time | optimization of both automatic and programmed parallelization. | | I would greatly appreciate hearing from anyone who can | provide advice on | an appropriate bibliography on the language implementation and past | work in this area (other than the code in the Haskell compilers and | libraries, of course), approaches, ongoing work, etc. | | Also, I would be most interested in hearing from anyone | interested in joining our effort (no problems with remote | access to the cluster by way of SSH), or making arrangements | for cooperation in research in this area, although, of | course, access to our equipment will be subject to approval | from "higher authorities" (however, note that I really don't | expect difficulty here, since our administration has already | shown considerable interest in supporting the project). | | Replies to the list will be read, but those who wish may mail | me directly at mgross@dorsai.org. | | Thanks in advance for any replies. | | Murray Gross | Adj. Lecturer, Brooklyn College, | City University of New York | | | | _______________________________________________ | Haskell mailing list | Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell |
-------------------------------------------------- Phil Trinder Department of Computing and Electrical Engineering Heriot Watt University Riccarton Edinburgh, EH14 4AS
E-mail: trinder@cee.hw.ac.uk Teleph: +44 (0)131 451 3435 Depart: +44 (0)131 451 3328 Fasmly: +44 (0)131 451 3327 Intrnt: http://www.cee.hw.ac.uk/~trinder
_______________________________________________ Haskell mailing list Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell
On Tue, 20 Aug 2002 paul@theV.net wrote:
Speaking of GpH, I wonder how is GdH coming along? It seems that the installation instruction on http://www.cee.hw.ac.uk/~dsg/gdh/ is still incomplete...
As far as I know, Mosix has nothing to do with PVM, so am I right to say that GpH still needs PVM even you use it on a Mosix cluster? Then what will Mosix provide when you can have a PVM cluster without Mosix running the same thing?
GpH needs PVM. The difference between PVM and mosix is that PVM requires the user to specify the processor on which tasks execute. I am beginning to investigate whether or not mosix will migrate PVM tasks, and if not, how this can be arranged. Mosix provides automatic load balancing, which is one of our major research interests. You are correct that PVM alone can be used for construction of a cluster; it is just that PVM clustering is not the kind of clustering we wish to achieve. And we may not even be able to get GpH to run on a non-shared-memory cluster. We may or may not be able to achieve our goals, but finding out is what research is all about, isn't it? :) Best, Murray Gross
Paul, The GdH implementation is currently used here, and is basically in beta-release form. If you would use it we could wrap up a and X86/Linux version for you. The GUM implementation of GpH and GdH uses very simple communication (mainly point to point, with a single barrier synchronisation, and the communication is restricted to just 3 C modules. As a result it can be readily ported to new communications libraries, e.g. MPI and CMMP (bespoke CM5 library) versions exist.
Then what will Mosix provide when you can have a PVM cluster without Mosix running the same thing?
Don't know a lot about Mosix, but if the PVM calls in the 3 GUM communications modules can be replaced with Mosix shared-memory routines that would give a low effort port of GUM to a Mosix cluster. I'd expect Mosix communication to be *much* faster than PVM: it's so generic that it absorbs a lot of time, e.g. probing for a message takes 2000 machine cycles. For small numers of processors a higher-performance implementation could be constructed by adapting the runtime system to effectively use the shared memory, but IIRC Simon Marlowe found there were problems with memory contention with more processors. Phil On Tue, 20 Aug 2002 21:17:28 +0800 paul@theV.net wrote:
Speaking of GpH, I wonder how is GdH coming along? It seems that the installation instruction on http://www.cee.hw.ac.uk/~dsg/gdh/ is still incomplete...
As far as I know, Mosix has nothing to do with PVM, so am I right to say that GpH still needs PVM even you use it on a Mosix cluster? Then what will Mosix provide when you can have a PVM cluster without Mosix running the same thing?
Regards, .paul.
On Tue, Aug 20, 2002 at 12:09:28PM +0100, Phil Trinder wrote:
Murray,
There are several parallel Haskell implementations: a survey of them has just appeared in Journal of Func. Prog Vols 4&5 (July & Sept 2002). Implementations are available for o Eden http://www.mathematik.uni-marburg.de/inf/eden/ o GpH http://www.cee.hw.ac.uk/~dsg/gph/
My group works on Glasgow parallel Haskell (GpH) which extends Haskell 98 with a parallel composition combinator. As Simon said the main implemntation, GUM, buys portability using the relatively slow PVM communications library. This doesn't matter so much on distributed memory machines, and we've recently achieved some quite respectable results SunServer shared-memory machines.
Simon Marlowe developed an alternative SMP implementation of GpH a couple of years ago that may be more suitable for a Mosix platform, but I'm not sure of the status of that implementation now.
Phil
On Mon, 19 Aug 2002 18:04:44 +0100 Simon Peyton-Jones <simonpj@microsoft.com> wrote:
Hi Murray
I'm catching up with my email backlog, but I didn't see a reply to your message, so I thought I'd reply. The parallel-Haskell crew are at http://www.cee.hw.ac.uk/~dsg/gph/ There's a mailing list that I'm ccing.
GHC can be built to run on a shared-memory multiprocessor, but we have not built and tested that for some time.
Untrue - both Eden and GpH are currently used on SunServer SMPs.
GPH is aimed more at distributed-memory machines, and has quite a bit more code layered on top of GHC. Details on the GPH page above.
Let me know if I can help. It'd be great to get some more parallel-Haskell stuff going.
Simon
| A group of us at Brooklyn College (City University of New | York) are in the latter stages of setting up a Mosix cluster | that we would like to use for research on, among other | things, parallel execution of Haskell code and run-time | optimization of both automatic and programmed parallelization. | | I would greatly appreciate hearing from anyone who can | provide advice on | an appropriate bibliography on the language implementation and past | work in this area (other than the code in the Haskell compilers and | libraries, of course), approaches, ongoing work, etc. | | Also, I would be most interested in hearing from anyone | interested in joining our effort (no problems with remote | access to the cluster by way of SSH), or making arrangements | for cooperation in research in this area, although, of | course, access to our equipment will be subject to approval | from "higher authorities" (however, note that I really don't | expect difficulty here, since our administration has already | shown considerable interest in supporting the project). | | Replies to the list will be read, but those who wish may mail | me directly at mgross@dorsai.org. | | Thanks in advance for any replies. | | Murray Gross | Adj. Lecturer, Brooklyn College, | City University of New York | | | | _______________________________________________ | Haskell mailing list | Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell |
-------------------------------------------------- Phil Trinder Department of Computing and Electrical Engineering Heriot Watt University Riccarton Edinburgh, EH14 4AS
E-mail: trinder@cee.hw.ac.uk Teleph: +44 (0)131 451 3435 Depart: +44 (0)131 451 3328 Fasmly: +44 (0)131 451 3327 Intrnt: http://www.cee.hw.ac.uk/~trinder
_______________________________________________ Haskell mailing list Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell
Haskell mailing list Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell
-------------------------------------------------- Phil Trinder Department of Computing and Electrical Engineering Heriot Watt University Riccarton Edinburgh, EH14 4AS E-mail: trinder@cee.hw.ac.uk Teleph: +44 (0)131 451 3435 Depart: +44 (0)131 451 3328 Fasmly: +44 (0)131 451 3327 Intrnt: http://www.cee.hw.ac.uk/~trinder
On Wed, Aug 28, 2002 at 10:46:34AM +0100, Phil Trinder wrote:
Paul,
The GdH implementation is currently used here, and is basically in beta-release form. If you would use it we could wrap up a and X86/Linux version for you.
That will be very sweet! I'd love to play with this thing ;-) Please post the link when it is ready.
...
Don't know a lot about Mosix, but if the PVM calls in the 3 GUM communications modules can be replaced with Mosix shared-memory routines that would give a low effort port of GUM to a Mosix cluster. I'd expect Mosix communication to be *much* faster than PVM: it's so generic that it absorbs a lot of time, e.g. probing for a message takes 2000 machine cycles.
That would be indeed a very good research field to work on, I guess Murray's group will be interested to pick it up from here. Though we are also experimenting some GpH stuff in a our company, we do not have the required research force or skill (yet). Regards, .paul.
On Wed, 28 Aug 2002 paul@theV.net wrote:
On Wed, Aug 28, 2002 at 10:46:34AM +0100, Phil Trinder wrote:
Paul,
The GdH implementation is currently used here, and is basically in beta-release form. If you would use it we could wrap up a and X86/Linux version for you.
That will be very sweet! I'd love to play with this thing ;-) Please post the link when it is ready.
A binary snapshot of GdH has been available since February, see the bottom of the GdH page: http://www.cee.hw.ac.uk/~dsg/gdh Note: best to install and test GpH first to verify that you have the required libraries... see ~dsg/gph cheers, ___ ][_) | +44 (0)131 451 3328 ]| \obert Pointon | http://www.cee.hw.ac.uk/~rpointon ------------------+------------------------------ Dept of Computing & Electrical Eng. Heriot-Watt University Riccarton Edinburgh, EH14 4AS
Robert, Thanks for this. Phil On Wed, 11 Sep 2002 15:20:03 +0100 (BST) "Robert F. Pointon" <rpointon@macs.hw.ac.uk> wrote:
On Wed, 28 Aug 2002 paul@theV.net wrote:
On Wed, Aug 28, 2002 at 10:46:34AM +0100, Phil Trinder wrote:
Paul,
The GdH implementation is currently used here, and is basically in beta-release form. If you would use it we could wrap up a and X86/Linux version for you.
That will be very sweet! I'd love to play with this thing ;-) Please post the link when it is ready.
A binary snapshot of GdH has been available since February, see the bottom of the GdH page: http://www.cee.hw.ac.uk/~dsg/gdh Note: best to install and test GpH first to verify that you have the required libraries... see ~dsg/gph
cheers,
___ ][_) | +44 (0)131 451 3328 ]| \obert Pointon | http://www.cee.hw.ac.uk/~rpointon ------------------+------------------------------ Dept of Computing & Electrical Eng. Heriot-Watt University Riccarton Edinburgh, EH14 4AS
To reduce the amount of duplication, I'm going to assume that everyone copied on your original note to me reads the list and copy only the list. If you know someone on the copy list who does not read the list, please tell me, and I will individually copy them on appropriate future e-mail. On Tue, 20 Aug 2002, Phil Trinder wrote:
Murray,
There are several parallel Haskell implementations: a survey of them has just appeared in Journal of Func. Prog Vols 4&5 (July & Sept 2002). Implementations are available for o Eden http://www.mathematik.uni-marburg.de/inf/eden/ o GpH http://www.cee.hw.ac.uk/~dsg/gph/
Yes, I have that survey, and am following up on it.
My group works on Glasgow parallel Haskell (GpH) which extends Haskell 98 with a parallel composition combinator. As Simon said the main implemntation, GUM, buys portability using the relatively slow PVM communications library. This doesn't matter so much on distributed memory machines, and we've recently achieved some quite respectable results SunServer shared-memory machines.
Although we have a multiprocessor machine available to us, we would prefer to concentrate our efforts on the cluster, because in these tight budget times, that seems to be a more likely avenue to increasing computational capacity.
Simon Marlowe developed an alternative SMP implementation of GpH a couple of years ago that may be more suitable for a Mosix platform, but I'm not sure of the status of that implementation now.
To the extent that we can use existing software, we will. All information on anything potentially useful is greatly appreciated. Thanks for your note. Best, Murray Gross
participants (4)
-
mgross＠dorsai.org -
paul＠theV.net -
Phil Trinder -
Robert F. Pointon